How bad is Gordon Hayward?

[Mike G]

I don't get to watch a lot of NBA, but when I have seen the Celtics, it seems they always do better when GH is not on the floor.
On this page --
https://www.basketball-reference.com/te ... 9/lineups/
-- we see the 20 most-used 5-man lineups, 4-man combinations, 3, and 2

What I have done is (1) Find Hayward in the 5-man units and (2) the other 4 players in the 4-man combos, then (3) deduce how the team has done in the remaining without-Hayward minutes those 4 other guys have played together.

There are 4 top-20 lineups that qualify:

Code: Select all

w/wo Hayward                wMin   MOV  woMin  MOV   diff  pts
Brown-Horford-Irving-Tatum   140  -3.8   134   6.0   -9.8  -29
Horford-Irving-Smart-Tatum    52 -10.0   341  17.6  -27.6  -30
Horford-Irving-Morris-Tatum   48  20.4   361  11.3    9.1    9
Horford-Irving-Morris-Smart   27  17.2   334  13.6    3.6    2

MOV is per 100 possessions. This is very close to per 48 minutes.
Diff is just MOV w - MOV wo
Pts is estimated by diff*wMin/48

Two big negatives and two small positives. Looking at other 5-man units containing that first other-4 sample, we note the Celts are:
2.6 pts better with Morris than with Hayward (65 min.)
19.4 pts/100 better with Smart (35 min)

For line 2 (H-I-S-T) we get:
18 pts better with Morris than with Hayward (303 min)
25 pts better with Brown (34 min)

Many of these minutes will be overlapping and describing the same conditions. So we are basically looking for any contraindications to the dismal analysis.
Three-man combinations with and without Hayward:

Code: Select all

. w/wo Hayward       wMin   MOV  woMin  MOV   diff   pts
Horford-Irving-Tatum  262   0.8   455  13.7  -12.9   -70
Brown-Horford-Irving  188  -5.5   196   6.9  -12.4   -48
Brown-Irving-Tatum    159  -0.3   198   7.1   -7.4   -25
Brown-Horford-Tatum   156  -4.7   186   4.7   -9.4   -30
Irving-Smart-Tatum    117  -3.1   489  18.8  -21.9   -54
Horford-Irving-Smart  112   5.3   394  15.9  -10.6   -25
Horford-Irving-Morris  94  11.2   433  10.1    1.1     2

Again, we don't add up all these lineups' net +/-, since they overlap. Even more so with the two-teammate combos:

Code: Select all

w/wo Hayward   wMin  MOV  woMin  MOV   diff   pts
Irving-Tatum   370   3.0   648  15.9  -12.9   -99
Horford-Irving 362   1.9   533  12.5  -10.6   -80
Brown-Rozier   356  -0.8   236   8.2   -9.0   -67
Horford-Tatum  323  -1.9   549  11.6  -13.5   -91
Brown-Irving   281  -1.6   311   2.0   -3.6   -21
Brown-Tatum    263  -2.9   320   8.6  -11.5   -63

This guy kills virtually every lineup he enters, but he still gets 26 mpg?

[Crow]

By RPM estimate, he has a -0.7 pts / 100 possessions negative impact. It might be an underestimate but the raw on / off has more differences than just him on / off. If same 4 data is bad for his big minutes lineups, he may be better in smaller lineups, perhaps where he is a main guy instead of a role player.

[Mike G]

The Celtics' SRS is 5.9, which would be achieved by 5 guys who average about 1.2 RAPM.
Here are their 1.5-season RAPM and current season raw on-off:

Code: Select all

Celts    rapm   raw
Tatum    3.66   5.2
Baynes   3.62   5.5
Irving   3.58   4.1
Theis    2.67   6.1
Morris   2.28   4.1
      
Hayward  1.66  -8.0
Smart    1.05   6.5
Brown    1.07  -5.4
Horford  0.44  -2.1
Rozier  -2.71  -9.7

Hayward didn't play last year, so his rapm is just this season.
RAPM of 1.66 is better than avg for the team, but his raw +/- is way low. It's consistent with the previous tables' higher-minutes rotations. For the other-4 lineups, the total points deficit divided by minutes he plays comes to -8.6 pts/48 min.
For the other-3 its -11; for other-2 its -10.3

Dink lineups, especially involving #11-14 guys, would seem the only explanation to raise his raw on-off to -8.
Are these mostly garbage minutes?

[schtevie]

I guess I don't understand some of the above calculations and data.

Looking at the four 4-man lineups first shown, I calculate the net points per 100 possessions differential (w/o Hayward - w/ Hayward) for each as:

B-H-I-T: +0.13 (that is, playing with any and all other players, this foursome scores, on net, 0.13 more points per 100 possessions better than when playing with Hayward)
H-I-S-T: +0.05
H-I-M-T: -0.16
H-I-M-S: +0.13

And a straight minute/possession weighted average of these differentials is +0.05

This suggests that Hayward has little negative effect comparatively, and this raw result is consistent with what one could expect from a player with a -0.77 RPM (relative to the ratings of his other teammates).

And then where does the Hayward raw -8.0 come from? NBA.com gives his +/- as +2.4.

[Mike G]

Code: Select all

w/wo Hayward                wMin   MOV  woMin  MOV   diff  pts
Brown-Horford-Irving-Tatum   140  -3.8   134   6.0   -9.8  -29
Horford-Irving-Smart-Tatum    52 -10.0   341  17.6  -27.6  -30
Horford-Irving-Morris-Tatum   48  20.4   361  11.3    9.1    9
Horford-Irving-Morris-Smart   27  17.2   334  13.6    3.6    2

MOV is per 100 possessions. This is very close to per 48 minutes.
Diff is just MOV w - MOV wo
Pts is estimated by diff*wMin/48

In 140 minutes of Brown, Horford, Irving, Tatum, and Hayward, the Celts were -3.8 pts/100 poss (as of a few days ago).
In 134 minutes of the first four and anyone but Hayward, they are +6.0/100

The difference is 9.8 pts/100 better with anyone but GH.
9.8*140/48 = 29 points deficit in those 140 minutes when he's on the floor.

With/without [Horford, Irving, Smart, and Tatum], the diff is 27.6 pts/100. Over 52 minutes, that's 30 points worse than alternatives.

If the Celts are +3 with Hayward and +11 without him, his net on-off is -8.0
He plays half time, and their overall MOV is +7 -- midway between 3 and 11.

Only 5 teams in history have had MOV>11. Highest by Boston is 2008, 10.3. Next best, 1986 was 9.4. Best Russell team, 1962, was 9.2

[Crow]

Need to be careful with terms. "Plus minus" is traditionally team plus minus with player on court. On / off may be stated as a plus, minus or neutral but it should not be called plus minus. It is on / off.

Hayward's plus minus is decent. His on / off is terrible.

[Mike G]

Yes.
Lakers with LeBron are +2.5, without him -4.0 -- On-minus-off = 6.5
When LAL and Bos play with both LeBron and Hayward in action, the teams are evenly matched.
When they both sit, the Celtics are 14.5 pts/100 better than LA.

[schtevie]

Alright, that's a wee bit embarrassing. Aside from having incorrectly transcribed the net effect for one lineup, I presented data as being in terms of per 100 possessions (well 48 minutes, as we are making that assumption) when what was shown was in terms of 1 possession. A factor of 100 mea culpa. I suppose the primary root of this carelessness was that Mike G's underlying argument seemed to me implausible and my mistake confirmed my priors.

This confessed, I revisit the issue: ought one believe that raw data should show GH to be quite bad when adjusted +/- data indicate him to be of "median to mean" quality? At this moment, RPM shows -0.77 (O:+0.20, D:-0.97) and BPM shows +0.2 (O:-0.5, D:+0.7).

Let me now present my corrected data for the four lineups in question, what reveals:

B-H-I-T: +12.9 (that is, playing with any and all other players, this foursome scores, on net, 12.9 more points per 100 possessions better than when playing with Hayward)
H-I-S-T: +5.1
H-I-M-T: -15.6
H-I-M-S: -0.4

And a straight minute/possession weighted average of these differentials is +5.0.

So, now I am obliged to scratch my head, as the Cs being +5.0 points per 48 minutes/100 possessions when playing a "non-Hayward" in these lineups suggests the possibility that RPM and BPM estimates might be misleading.

But delving a bit further into the issue, I don't think that referring to GH's contributions to the most used lineups provides any evidence for this conjecture.

Basically, there seems to have been an oversight of 6 of the top-10 GH lineups shown at the provided link (showing the top-20 Celtics lineups by MP). Chosen are the 2nd, 6th, 9th, and 19th. For whatever reason, not included are the 5th, 7th, 11th, 16th, 17th, and 18th most played lineups.

Doing the same calculation (hopefully correctly this time) for just these excluded lineups, what I find is that the minute weighted average of the "GH vs. all other Celtics" is -12.6. That is, Gordon's contribution is assumed to be much higher than that of his replacements.

Then pooling all the top-10 lineups together, we get -3.4 points per 48 minutes/100 possessions.

So, in raw terms, for the top-10 lineups, what are 41% of his minutes played, his contribution is greater than that of his replacements, and by a pretty significant amount.

Should this be surprising? I still say "no". Is it determinative? No, for all the obvious caveats: teammates' contributions aren't the relevant baseline, it isn't controlling for the quality of opponents, and all that.

Setting this aside, there is the underlying (overarching?) issue of the apparent deterioration in GH's productivity compared to his Utah years, what seems pretty real. What might explain that? Well, the injury, of course. Then adapting to a new system and new teammates, amongst a host of other possible explanations.

For fun, given that we are focused on lineup data, I went and looked at Utah lineup data, and one thing that struck me is a fundamental difference between how the 2016-17 Jazz and the 2018-19 Celtics manage court time. Specifically, the top-10 GH Jazz line-ups had an average MP/game of 6.1 minutes (or 11.9 possessions given their pace). By contrast, the top-10 GH Celtics line-ups only run for an average of 3.6 MP/game (or about 7.5 possessions per game).

I don't know if that might explain something or not, but it is just a nice reminder of how insane the NBA is when it comes to substitutions.

[Mike G]

B-H-I-T: +12.9 (that is, playing with any and all other players, this foursome scores, on net, 12.9 more points per 100 possessions better than when playing with Hayward)...

https://www.basketball-reference.com/te ... 9/lineups/
The most-used lineup with Hayward includes Brown, Horford, Irving, and Tatum; they have 139.7 minutes together.
That's almost exactly half of the total minutes (276.9) of that other 4 together (w + wo GH)

So roughly speaking, in half of BHIT minutes -- with Hayward -- they are -3.8 pts/100
In all BHIT minutes, they are +0.6
Therefore, in the other 138 minutes -- with a 5th player who is not Hayward -- they are 3.8 pts better than 0.6, or +4.4
The with-minus-without difference is (4.4+3.8) 8.2. How do you get 12.9 ?

[schtevie]

The calculation of the -12.9 counterfactual is as follows.

All B-H-I-T lineups (prior to tonight's game) were a +11 net points in 276 minutes, what implies 1.91 net points per 48 minutes (approximate 100 possessions).

The GH-B-H-I-T lineup was a -13 net points in 139.8 minutes, correspondingly a -4.5 net points per 48 minutes.

This implies that non-GH-B-H-I-T lineumps were +24 and 137.1 minutes and +8.4 net points per 48 minutes.

Subracting +8.4 from -4.5, you get the relative producivity of non-GH lineups vs. GH. That is you "naively" expect to gain 12.9 points per 48 minutes/approximate possessions by utilizing the actual "average non-GH addition" to the B-H-I-T foursome.

Then the weight of this particular lineup in an averaging calculation is, of course, the 139.8 minutes of the GH lineup.

Make sense?

[Mike G]

All B-H-I-T lineups (prior to tonight's game) were a +11 net points in 276 minutes, what implies 1.91 net points per 48 minutes (approximate 100 possessions).

The GH-B-H-I-T lineup was a -13 net points in 139.8 minutes, correspondingly a -4.5 net points per 48 minutes.

Where do you find these "net points"?
The table on the linked page shows "Net (Per 100 Poss) Pts" as +0.6 for 277 Brown-Horford-Irving-Tatum minutes.
And in the 5-man combos, adding Hayward, that number is -3.8
Not including last night's game.

[schtevie]

In the general lineup finder (not via the team pages) the default sort is Net Points.

[Mike G]

Wow, thanks for that. Big difference in what they make available.
You were converting from net points to per48, and I was going the opposite way.
http://bkref.com/tiny/Tmjcm
On the team page, showing only top 20 most used 5-man units, 4-man combos, etc, I could only find a handful that appear in 2 adjacent tables. Now we have the whole spread, including insignificant minutes.

There are 13 lineups with Hayward for >20 minutes; of these, 10 have same-other-4 for 20+ min.
From the lineup finder page, we get minutes and net points for the 4-man combo; minutes and net pts with Gordon Hayward (on 5 man page). Then just subtract to get without GH MP and Net

Code: Select all

w/wo GH  MP  Net   wMP  Net   woMP  Net
BHIT    280    8   140  -13    140   21
BRTTh    95   10    71   20     24  -10
HIST    456  128    59   15    397  113
HIMT    468  117    49   29    419   88
HRST    165   21    41    1    124   20

HIMS    421   94    32    8    388   86
ISTTh    71   -3    28   -1     43   -2
BaBIR    47    3    24    0     22    3
BORTh    43    4    20   -2     23    6
BOThW    40   20    20   15     20    5

top10  2086  402   484   72   1601  330
per48   all  9.3   wGH  7.1   woGH  9.9

That bottom line shows with-and-without Hayward for the sum of minutes here, and the difference is only -2.8 pts/48. This is probably in line with most of the league's 6th/7th men who sometimes start.

This still only captures 484, or 3/8, of Hayward's 1289 minutes this year. Is he a dink lineup specialist?
Here's the same format for 3-person combinations with and without GH:

Code: Select all

w/wo    MP - Net    wMP - Net   woMP - Net
HIT    786   166    269    33    518   133
BHI    393    -7    190   -24    203    17
BIT    370    31    160    -3    210    34
BHT    372     3    159   -13    213    16
HIS    577   141    127    28    450   113

IST    672   206    124    20    548   186
BRT    251    14    120    17    131    -3
BMR    225    24    118    19    108     5
HST    623   133    107    18    516   115
HIM    591   113    100    27    491    86

top10 4861   824   1474   122   3387   702
per48  all   8.1    wGH   4.0   woGH   9.9

The top 10 are combos with and without Haywood for 100+ minutes. Now his on-off is -5.9
Finally, hoping for some closure:

Code: Select all

w/wo    MP - Net    wMP - Net   woMP - Net
IT    1097   278    380    53    717   225
HI     975   164    379    36    596   128
HT    1016   153    356    19    659   134
BT     649    74    295    18    353    56
RT     633    20    272   -19    361    39

BH     539   -21    243   -39    295    18
IS     886   199    234    46    651   153
ST    1015   251    225    53    790   198
MR     560    -3    224   -37    336    34
HS     784   138    195    22    589   116

10    8151  1253   2803   152   5349  1101
per48  all   7.4    wGH   2.6   woGH   9.9

Seems all of these combos are worse with Haywood than with others. On-off of -7.3 is approaching replacement level perhaps?

[Crow]

The big picture summary of this case so far is he can have positive plus minus but there are much better alternatives / results with him off.

The next step imo is look at splits of the season. His plus minus was moderately positive last fall, moderate bad in January and now great in early Feb. He has shot much better in the new year.

Looking even closer I see an especially bad 9 game stretch from Jan. 10-26 where he was -67 on the court. Outside that he is plus 200 for rest of season. This was a couple weeks after the shift to the current starting lineup. If it was a funk from being demoted, it was a delayed reaction. During this stretch Stevens did not use him in any lineup for an average of 2 minutes per game. Most of these lineups did poorly. Random small lineups is not good usage. If I had more interest- or if either of you do- the next step would probably be to compare lineup usage and results before Jan. 10-26, then and after. That period started with 3 road losses but then returned to mostly at home, mostly against eastern teams and a not particularly hard mix of opponents.


In the 6 games after this bad period, he has 2 lineups used over 2 minutes / gm with one almost at 4 min. Small progress. Most lineups with Rozier. One really strong one with Rozier Brown Tatum Theis. It is actually his 2nd most used lineup for season though at a paltry 70 minutes or an average of about 1.2 minutes per game. In the most recent 6 games that is up to a wopping 2.2 min / g. With 4 non-starters this would be comparatively easy to play more, a lot more to test, prove or have regress back to mean.

Back to season level data for a minute. Hayward has 5 strong lineups, 4 mildly positive, 1 negative. Here is an outsider's thought: play the really strong ones more. It isn't all about him and there are lineups better without him but within the constraints of overall rotation and player minute limits, play the best performing Hayward lineups more than the lesser ones. The insiders aren't. Somebody ask them why not? Not that they'd answer. I've tried.

If you might say that almost all these lineup are really small sample and quality can not be assessed in a statistically significant way, I'd say true... SO PLAY FEW LINEUPS MORE, THE BEST ONES TO DATE AND LESS RANDOM DINK LINEUPS. But that is just me. The "world's best coach" and his assistants prefer to use all sorts of dink lineups presumably selected with brilliance but somehow mostly fail to be positive.

[Crow]

Actually this strong Hayward lineup only got used twice in last 6 games. Once briefly and then 12 great minute in last game. Was this just a random big minute game and then to return to mostly ignoring it or the sign of a strategy change (consistent with my recommendation above)? We'll see. For 2/3rds of the season they didnt do this.